Stopping and starting BOINC is wasting time on a multi-core cruncher. Simply disable LAIM (fka Leave Application In Memory, now Leave Non-GPU tasks in memory), then suspend the task that is stuck, wait 30 seconds to clear memory [look in event log or Task Manager], then resume it. BEFORE doing this, suspend all Ready to Start tasks (RtS) and disable network [activity menu], so the client does not go out and fetch new work or start it while you cycle the stuck task. This way all other running tasks are not affected, which can save you hours of computing time, depending on how frequent they checkpoint progress.

I get it more on this project than on any other. I had one work unit that got stuck at least twice.

It can be frustrating if you're not paying attention and therefore don't see that it's got stuck until it's racked up several hours of fruitless computing. But I'd echo others' comments that you don't have to abort the task, just shut BOINC and start it again, although I guess that would be problematic if you were crunching another project with long gaps between checkpoints.

ETA: I've one work unit right now that froze at 4% and has now frozen at 8%. If it keeps doing this, I may just abort it, notwithstanding what I said above - have to keep waiting for the other project's checkpoint before I can restart the client. Annoying.

Go ahead and abort it if you are able. It will get re-issued. If it is a bad unit , it will get about 5 or 6 iterations and then get pulled completely.
Cheers